MindLab-UNAL: Comparing Metamap and T-mapper for Medical Concept Extraction in SemEval 2014 Task 7

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 424–427, Dublin, Ireland, August 23-24, 2014. MindLab-UNAL: Comparing Metamap and T-mapper for Medical Concept Extraction in SemEval 2014 Task 7 Alejandro Riveros, Maria De-Arteaga, Fabio A. Gonz´alez and Sergio Jimenez Universidad Nacional de Colombia Ciudad Universitaria Bogot´a, Colombia [lariverosc,mdeg,fagonzalezo, sgjimenezv]@unal.edu.co Henning M¨uller Univ. of Applied Sciences Western Switzerland, HES-SO Sierre, Switzerland henning.mueller@hevs.ch Abstract This paper describes our participation in task 7 of SemEval 2014, which focuses on analysis of clinical text. The task is divided into two parts: recognizing men- tions of concepts that belong to the UMLS (Unified Medical Language System) se- mantic group disorders, and mapping each disorder to a unique UMLS CUI (Concept Unique Identifier), if possible. For identi- fying and mapping disorders belonging to the UMLS meta thesaurus, we explore two tools: Metamap and T-mapper. Addition- ally, a Named Entity Recognition system, based on a maximum entropy model, was implemented to identify other disorders. 1 Introduction Clinical texts are unstructured data that, when pro- cessed properly, can be of great value. Extracting key information from these documents can make medical notes more suitable for automatic pro- cessing. It can also help diagnose patients, struc- ture their medical histories and optimize other clinical procedures and research. The task of identifying mentions to medical concepts in free text and mapping these mentions to a knowledge base was recently proposed in ShARe/CLEF eHealth Evaluation Lab 2013, at- tracting the attention of several research groups worldwide (Pradhan et al., 2013). The task 7 in SemEval 2014 (Pradhan et al., 2014) elaborates in that previous effort focusing on the recognition and normalization of named entity mentions be- longing to the UMLS semantic group disorders. The paper is organized as follows: in section 2 we briefly present the data, section 3 contains the This work is licensed under a Creative Commons At- tribution 4.0 International Licence. Page numbers and pro- ceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ description of the methods and tools used in our system. Later, on sections 4 and 5 we provide the details of the three submitted runs and expose the official results. Finally, sections 6 and 7 include discussions on variations that could be done to im- prove performance and conclusions to be drawn from our participation in the task. 2 Data Description The training data for SemEval 2014 Task 7 con- sists of the ShARe (Shared Annotation Resource) corpus, which contains clinical notes from MIMIC II database (Multiparameter Intelligent Monito– ring in Intensive Care). The data were manually annotated for disorder mentions, normalized to a UMLS Concept Unique Identifier when possible, and marked as CUI-less otherwise. Four types of reports where found in the cor- pus: 61 discharge summaries, 54 ECG reports, 42 ECHO reports and 42 radiology reports, for a to- tal of 199 training documents, each containing se– veral disorder mentions. 3 Methods Used 3.1 Named-Entity Recognition Using the Java libraries Apache OpenNLP1 and Maxent2, a maximum entropy model …

Content truncated to 500 words